In a dense small cell deployment scenario, users are always prone to suffer severe interferences from neighbor base stations (BS) because the BSs are usually located closely. Coordinated Multi-Point (CoMP) can be introduced to alleviate these interferences and improve the system performance. It is necessary to determine coordination areas (CA) before implementation. In this paper, a novel dynamic clustering algorithm in CoMP joint transmission system is proposed based on graph theory. Firstly a feedback procedure is designed for interference reports mainly based on large scale fading. By building a graph according to the interferences, the clustering problem is equivalent to dividing the graph into several subgraphs. Each subgraph represents a CoMP cluster. It can be solved through a greedy strategy that each BS searches its best coordinated BSs. Compared with some other dynamic algorithms, the complexity of the proposed scheme is lower because it can be implemented in a decentralized way. Therefore this method is suitable in dense cell deployment with a large number of BSs. The simulation results show that the novel clustering algorithm performs better in user capacity than other traditional dynamic schemes. The influences of some parameters in this method are also considered and evaluated in the simulation.